New single-microphone localization technique beats multi-mic methods

A new technique from Matthew Maciejewski, published at IEEE ICASSP 2026, solves audio source localization using just one microphone. Traditional methods rely on multiple microphones to triangulate a source's position using time-of-arrival differences. This work flips the problem: it uses the reverberation tail—the late part of sound reflections that depends only on the room, not on the exact source position. That invariance provides a stable reference to compare signals.

The method applies Weighted Prediction Error (WPE) dereverberation to estimate the late tail, then uses a probabilistic framework to compute the likelihood that two audio clips came from the same location. Testing on speaker diarization tasks shows accuracy rivaling multi-microphone systems. This means future smart speakers, meeting transcribers, or hearing aids could locate speakers without a complex array, using only a single mic and clever signal processing.